Towards Data Science

This account is verified. Learn more

@TDataScience

The world's leading publication for data science and artificial intelligence professionals.

Submit an Article ✍️ contributor.insightmediagroup.io

We're Global 🌏 towardsdatascience.com Joined October 2016

1.9K Following 252K Followers 65K Posts

Latest posts 5

Towards Data Science @TDataScience 3h

How much of vLLM’s speedup is actually the scheduler? @nhu_hoang isolated it: continuous batching delivered up to 2.91× over static batching, but exactly 1.00× when every reply had the same length. It all comes down to how much your output lengths vary. towardsdatascience.com/continuous-bat…

towardsdatascience.com Continuous Batching Gave 1.00× at Zero Variance and 2.91× on Chat-Like Traffic For anyone running an inference server: a zero-variance control, five traffic traces, and an engine-vs-scheduler split show where the gains come from.

Towards Data Science @TDataScience 5h

A hundred people ask the same AI for ideas. Each gets a better answer than they’d have written alone, and the room ends up with nearly the same answer a hundred times. @DattijoMuhammad weighs the evidence on AI and creativity, and suggests a simple fix. towardsdatascience.com/ask-ai-to-refi…

towardsdatascience.com Writers Got Better With AI Ideas. Their Stories Got More Alike. Studies of short stories, metaphors, and 486 participants point one way, an 800-person experiment points the other, and the author weighs both.

Towards Data Science @TDataScience 7h

Two maps found 1,515 forests. A fisheries formula from the 1890s said about 207 were missing from both. Maksym Bilychenko applies the same trick to AI agents, estimating what your eval never sees, with only a 10% hand audit. towardsdatascience.com/how-to-estimat…

towardsdatascience.com Two Runs of One LLM Agreed So Well They Underestimated the Total by 22% A fisheries estimator from the 1890s, tested on 294 CUAD contracts, landed within 2 points of true recall for most agents with a 10% hand audit.

Towards Data Science @TDataScience 8h

OLS predicted $164K for a house that sold for $219K. Quantile regression gave a range, $136K to $227K, that contained it. @gurezende shows how to build prediction intervals that widen where your data gets noisier. towardsdatascience.com/quantile-regre…

towardsdatascience.com OLS Missed This House Price by $55K. Quantile Regression Didn't. On 88 house sales, OLS predicted $164K for a $219K home, while a 90% interval from 5th and 95th percentile models spanned $136K to $227K.

Towards Data Science @TDataScience 12h

Most multi-agent systems come down to two moves: call a specialist agent as a tool, or hand off control entirely. Shuai Guo builds both in Python with the OpenAI Agents SDK, then shows how fan-out, pipelines, hierarchies, and swarms reuse them. towardsdatascience.com/how-to-build-m…

towardsdatascience.com How to Build Multi-Agent Systems with the OpenAI Agents SDK Build an EV purchase adviser and a support agent in Python, then see how fan-out, pipelines, hierarchies, and swarms reuse the same building blocks.